J
OURNAL DE
T
HÉORIE DES
N
OMBRES DE
B
ORDEAUX
P
IETER
M
OREE
P
ETER
S
TEVENHAGEN
Prime divisors of the Lagarias sequence
Journal de Théorie des Nombres de Bordeaux, tome
13, n
o1 (2001),
p. 241-251
<http://www.numdam.org/item?id=JTNB_2001__13_1_241_0>
© Université Bordeaux 1, 2001, tous droits réservés.
L’accès aux archives de la revue « Journal de Théorie des Nombres de Bordeaux » (http://jtnb.cedram.org/) implique l’accord avec les condi-tions générales d’utilisation (http://www.numdam.org/conditions). Toute uti-lisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit conte-nir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
Prime divisors of the
Lagarias
sequence
par PIETER MOREE et PETER STEVENHAGEN
RÉSUMÉ. Nous donnons une solution à un
problème posé
parLagarias
[5]
en 1985, en déterminant sous GRH la densité del’ensemble des nombres premiers qui sont des diviseurs de
ter-mes de la suite définie par x0 = 3, x1 = 1 et la relation
de récurrence xn+1 = xn + xn-1. Cela donne le premier
exem-ple
d’une suite de récurrence d’ordre 2 qui n’est pas ’à torsion’ pourlaquelle
on sait déterminer la densité associée des diviseurspremiers.
ABSTRACT. We solve a 1985
challenge problem posed by
La-garias
[5]
by
determining,
underGRH,
thedensity
of the set ofprime numbers that occur as divisor of some term of the sequence defined
by
the linear recurrence xn+1 = xn + xn-1 andthe initial values x0 = 3 and x1 = 1. This is the first
example
ofa ’non-torsion’ second order recurrent sequence with irreducible
recurrence relation for which we can determine the associated
den-sity of prime divisors.
1. Introduction
In
1985,
Lagarias
[5]
showed that the set ofprime
numbers that dividesome Lucas number has a natural
density
2/3
inside the set of allprime
numbers. Here the Lucas numbers are the terms of the second orderrecur-rent sequence defined
by
the linear recurrence = xn + xn-1and the initial values xo = 2 and
xl = 1.
Lagarias’s
method is aquadratic
analogue
of theapproach
usedby
Hasse[2, 3]
indetermining,
for agiven
non-zerointeger
a, thedensity
of the set of theprime
divisors of thenum-bers of the form an + 1. Note that the sequence +
1 loo 1
also satisfiesa second order recurrence.
Hasse and
Lagarias apply
the Chebotarevdensity
theorem to a suitabletower of Kummer fields. Their method of ’Chebotarev
partitioning’
canbe
adapted
to deal with the class of second order recurrent sequences thatare now known as ’torsion
sequences’
[1,
8,
11].
For second order recurrentinteger
sequences that do notenjoy
the ratherspecial
condition ofbeing
’torsion’,
it can nolonger
beapplied.
In the case of the the Lucasnumbers,
changing
the initial values into xo = 3 andxl = 1
(while
leaving
therecur-rence xn+1 = xn + xn-1
unchanged)
leads to a sequence for whichLagarias
remarks that his method
fails,
and he wonders whether some modificationof it can be made to work.
We will
explain
how non-torsion sequences lead to aquestion
that isreminiscent of the Artin
primitive
rootconjecture.
Inparticular,
we will see that for a non-torsion sequence, there is no number field F(of
finitedegree)
with theproperty
that allprimes
having
agiven
splitting
behav-ior in F divide some term of the sequence. It follows that
Chebptarev
partitioning
can not beapplied directly.
However,
it ispossible
tocom-bine the
technique
of Chebotarevpartitioning
with theanalytic techniques
employed by Hooley
[4]
in hisproof
(under
assumption
of thegeneralized
Riemannhypothesis)
of Artin’sprimitive
rootconjecture.
In the case ofthe modified Lucas sequence
proposed by Lagarias,
wegive
a fullanalysis
of the situation and prove the
following
theorem.Theorem. Let the
integer
sequencedefined by
xo =3,
xl = 1and the linear recurrence + xn-l- ·
If
thegeneralized
Riemannhypothesis
holds,
then the setof
prime
numbers that divide some termof
this sequence has a natural
density.
Itequals
Numerically,
one finds that 45198 out of the first 78498primes
below106
divide the sequence: a fraction close to .5758.
2. Second order recurrences
Let X = second order recurrent
sequence. It is our aim to
determine,
whenever itexists,
thedensity
(inside
the set of allprimes)
of the set ofprime
numbers p that divide some term of X.We let = be the recurrence satisfied
by
X,
and denoteby f
=T 2 -
ao EZ[T]
thecorresponding
characteristicpolynomial.
We factorf
over analgebraic
closure ofQ
asf
=(T -
a)(T -
a).
In order to avoid
trivialities,
we will assume that X does notsatisfy
afirst order recurrence, so that aa =
ao does not vanish. The root
quotient
r =
r( f )
of the recurrence, which isonly
determined up toinversion,
isthen defined as r =
a/a.
It is either a rational number or aquadratic
irrationality
of norm 1. In theseparable
caser ~
1 we haveAs our sequence is
by
assumption
not of order smaller than2,
we have0. Denote
by
the initial
quotient q
=q(X)
of X. Just as the rootquotient,
this is anum-ber determined up to inversion that is either rational or
quadratic
of norm 1.The
elementary
but fundamental observation for second order recurrencesis that for almost all
primes
p, we have the fundamentalequivalence
p divides xn 4=~’
-c/c
=(a/a)n
EO/pO
Here 0 is the
ring
ofintegers
in the fieldgenerated by
the roots off .
This is thering
Z iff
has rational roots, and thering
ofintegers
of thequadratic
field(a ~X ~ / ( f )
= otherwise. Theequivalence
above doesnot make sense for the
finitely
manyprimes
p for which either r or q is notinvertible modulo p, but this is irrelevant for
density
purposes.In the
degenerate
case where the rootquotient
r is a root ofunity,
it iseasily
seen that the set ofprimes
dividing
some term of X is either finiteor cofinite in the set of all
primes.
We will further exclude this case, which includes theinseparable
case r =1,
forwhich q
is not defined.As we are
essentially
interested in the set ofprimes
p forwhich q
is in thesubgroup generated by
r in the finite group of invertibleresidue classes modulo p, we can formulate the
problem
we aretrying
tosolve without any reference to recurrent sequences.
Depending
on whetherthe root
quotient
r is rational orquadratic,
this leads to thefollowing.
Problem 1. Given two non-zero rational numbers q andr ~
f1,
compute,
whenever it
exists,
thedensity of
the setof
primes
pfor
which we haveq
mod p
E(r
mod p)
cFJ§ .
Problem 2. Let r be a
quadratic
irrationality of
norm 1 and C~ thering
of
integers
of Q (r).
Given anedement q
EQ (r)
of
normI,
compute,
wheneverit
exists,
thedensity
of
the setof
rationalprimes
pfor
which we haveq
mod p
E(r
mod p)
c(O/pO)*.
The instances of the two
problems
above where(q
modr)
is a torsionele-ment in the group
Q(r)*/(r)
are referred to as torsion cases of theprob-lem,
and the sequences thatgive
rise to them are known as torsionse-quences. The sequences + studied
by
Hasse,
the Lucas sequence +6-nloo o
with E =1+2 treated
by Lagarias
and theLucas-type
se-quences in
[8]
aretorsion;
infact,
they
allhave q
= -1. The main theorem for torsion sequences, for which we refer to(11~,
is thefollowing.
Theorem. Let X be a second order torsion sequence. Then the set 1rx
of
prime
divisorsof
X has apositive
rationaldensity.
3. Non-torsion sequences
Problem 1 in the
previous
section is reminiscent of Artin’s famousquestion
on
primitive
roots:given
a non-zero rational numberr ~
~1,
for howmany
primes
p does rgenerate
the groupFp
of units modulop?
(One
naturally
excludes thefinitely
manyprimes
pdividing
the numerator ordenominator of r from
consideration.)
Artin’sconjectural
answer to thisquestion
is based on the observation that the index[Fp* :
(r)]
is divisibleby
j
if andonly if p splits completely
in thesplitting
fieldFj =
Q((j,
rl/j)
of thepolynomial
Xi -
r overQ. Thus,
r is aprimitive
root modulo p if andonly
p does notsplit completely
in any of the fieldsFj
with j
> 1. For fixedj,
the setSj
ofprimes
that dosplit completely
inFj
has naturaldensity
1/[Fj :
Q]
by
the Chebotarevdensity
theorem.Applying
aninclusion-exclusion
argument
to the sets8j,
oneexpects
the set S =S, B
ofprimes
for which r is aprimitive
root to have naturaldensity
J-.
Note that the
right
hand side of(3.1)
converges for all r EQ* B {±1}
as[Fj :
Q]
is a divisor ofcp ( j ) ~ j
with cofactor boundedby
a constantdepending
only
on r.A
’multiplicative
version’ of the ’additive formula’(3.1)
for5(r)
is ob-tained if one starts from the observation that r EQ* B
is aprimitive
root if and
only
if P
does notsplit completely
in any fieldFe
with iprime.
The fieldsFi
are ofdegree
~(~ 2013 1)
for almost allprimes
andusing
thefact that
they
are almost’independent’,
one cansuccessively
eliminate theprimes
thatsplit completely
insome FE
to arrive at a heuristicdensity
The correction factor cr for the
’dependency’
between the fieldsFt
isequal
to 1 if thefamily
of fields islinearly
disjoint
overQ,
i. e., if each fieldis
linearly disjoint
overQ
from thecompositum
of the fieldsFe
witht 0
fOe
If r is not aperfect
power inQ*,
we have =cr .
It turns out that the
only possible
obstruction to the lineardisjointness
of the fieldsFt
occurs whenF2
= isquadratic
of odd discriminant.In this case,
F2
is contained in thecompositum
of the fieldsFt
with tdividing
its discriminant. The value of cr is a rationalnumber,
and one canFor
example,
taking
r =5,
one hasF2
CF5
and thesuperfluous
’Eulerfactor’ 1 -
(F5 :
19
at t = 5 in theproduct
IIl(1 -
(Fe :
Ql-’)
is ’removed’by
the correction factor c5 =a 5
=20
It is non-trivial to make the heuristics above into a
proof.
AsHooley
[4]
showed,
it can be done if one iswilling
to assume estimates for theremainder term in the
prime
number theorem for the fieldsFj
that arecurrently only
known to hold underassumption
of thegeneralized
Riemannhypothesis.
One should realize thatonly
when we considerfinitely many t
(or j)
at atime,
the Chebotarevdensity
theoremgives
us the densities wewant. After
taking
a ’limit’ over all~,
weonly
know that theright
handside of
(3.1)
or(3.2)
is an upperdensity
for the set ofprimes
p for whichr is a
primitive
root. We have however noguarantee
that we are left with anon-empty
set of such p. Put somewhatdifferently,
we can not obtainprimes p
for which(r
modp)
is aprimitive
rootby
imposing
asplitting
condition on p in a number field F of finite
degree; clearly,
there isalways
some fieldFe
that islinearly disjoint
fromF,
and nosplitting
condition inF will
yield
the ’correct’splitting
behavior inFt.
A similarphenomenon
occurs in the
analysis
of non-torsion cases of the Problems 1 and 2. This isexactly
what makes non-torsion sequences so much harder toanalyze
than torsion sequences.If
(q mod r)
is not a torsion element in(a*/(r),
then Problem 1 canbe treated
by
ageneralization
of thearguments
usedby
Artin. For eachinteger i > 1,
one considers the set ofprimes
p(not
dividing
the numeratoror denominator of either q or
r)
for which the index~FP :
(r)]
isequal
toi and the index
[Fp*
(q)]
is divisibleby
i. These are theprimes
thatsplit
completely
in the fieldFi~l
=Q( (i,
rl/i,
ql/i),
but not in any of the fieldsFi,j =
with j
> 1. Asbefore,
inclusion-exclusionyields
aconjectural
value for thedensity
8i(r,
q)
of this set ofprimes,
andsumming
over i we
get
as a
conjectural
value for thedensity
in Problem 1. Note that isnothing
but theprimitive
rootdensity
6(r)
from(3.1).
The condition that
(q
modr)
is not a torsion element in(a*/(r)
meansthat q
and r aremultiplicatively independent
inQ*.
In this caseQ]
is a divisor of
i2j ~
cp(ij)
with cofactor boundedby
a constantdepending
only
on q and r. Thus the double sum in(3.3)
converges, and under GRH one can show[9, 10]
that its value is indeed thedensity
one is asked todetermine in Problem 1.
As in Artin’s case, one can obtain a
multiplicative
version of(3.3)
by
ap)
C(r
modp)
inFp
means that for allprimes
we have an inclusion(3.4)
(q
modp) g
C(r
modp) g
of the
~-primary
parts
of thesesubgroups.
If we fix both t and the numberk =
ordg(p -
1)
of factors t in the order ofFp,
this condition can berephrased
in terms of thesplitting
behavior of p in the number fieldMore
precisely,
we haveordP(P - 1)
= k if andonly
if psplits completely
in
Q(Cfk)
but not inQ((tk+1);
of theprimes
p that meet thiscondition,
we want those p for which the order of the Frobenius elements over p in
divides the order of the Frobenius elements over
p in
By
the Chebotarevdensity
theorem,
onefinds that the set of
primes
p withorde(p -
1)
=k,
which hasdensity
~-k
for k >
1,
is a union of two sets that each have adensity:
the set ofprimes
p for which the inclusion
(3.4)
holds and the setof P
for which it does not.This ’Chebotarev
partitioning’
allows us tocompute,
for each~,
thedensity
of theprimes
p for which we have the inclusion(3.4):
summing
over k in theprevious
argument
yields
a lowerdensity,
and this is therequired
density
as we canapply
the sameargument
to thecomplementary
set ofprimes.
For all but
finitely
many~,
the fieldsQ((tk+1,ql/tk)
andQ((tk+1,rl/tk)
are
linearly disjoint
extensions of with Galois group for allk > 0. In this case the set of
primes
p withordP(P -
1)
= kviolating
(3.4)
has
density
Summing
overk,
we find that(3.4)
does not hold for a set ofprimes
ofdensity
~/(~3 -
1).
As the fieldsfor
prime
values of £ form alinearly disjoint family
if we excludefinitely
many ’bad’
primes t,
themultiplicative analogue
of(3.3)
readsAs is shown in
[9],
the ’correction factor’ cq,r is a rational number thatadmits a somewhat involved
description
in termsof q
and r. Inpractice,
one finds its value most
easily by
starting
from the additive formula(3.3) .
In the situation of Problem2,
thearguments
just given
can be taken overthose rational
primes
p thatsplit
completely
in 0.Writing
K =Q(r)
andwe find that the
density
of the rationalprimes
p that aresplit
in 0 and forwhich we
have q
mod p
E(r
mod p)
Cequals
as in the case of Problem
1,
one needs to assume thevalidity
of the gen-eralized Riemannhypothesis
for this result.By
(3.7),
thecomputation
ofbsplst
amounts to adegree
computation
for thefamily
of fields Infact,
because of the numerator in(3.7)
one may restrict to the casewhere j
issquarefree.
As in the case of Problem1,
one finds(under GRH)
that the
split density equals
L
for some rational number
cir.
The next sectionprovides
atypical
example
of such a
computation.
It shows that the value ofcir
is not assimple
afraction as the
analogous
factor cr
in(3.2).
For the rational
primes
p that are inert in0,
the determination of thecorresponding density
>inert
is more involved than in thesplit
case. Thegroup in Problem 2 is now
cyclic
of orderp2 -
1,
and(q
modp)
and
(r
modp)
are elements of the kernelof the norm map, which is
cyclic
of orderp+1.
In order to have the inclusion(3.4)
ofsubgroups
of Kp for allprimes
we fix £ and k =orde(p
+1) >
1 andrephrase
(3.4)
in terms of thesplitting
behavior of p in thequadratic
counterpart
B-’-7 1 /-t ,
of
(3.5).
Let us assume forsimplicity
that t is an oddprime,
and thatK is not the
quadratic
subfield ofQ((i).
Then therequirement
that p be inert in K andsatisfy
ordt(p
+1)
= k > 1 means that the Frobeniuselement
of p
inGal(K((tk+1)/Q)
is non-trivial on K and has order 2£ when restricted to LetBk
C be the fixed field of thesubgroup
generated by
such a Frobenius element. ThenBk
does not contain K or(a(~e),
andBk
CK((tk)
is aquadratic
extension. Let ak be the non-trivialautomorphism
of this extension. Then ak actsby
inversion on andthe norm-1-condition on q and r means that ak also acts
by
inversion on q and r. The Galoisequivariancy
of the Kummerpairing
shows that the natural action of Qk on is
trivial,
so is abelian overBk.
It is thelinearly disjoint compositum
ofthe
cyclotomic
extensionBk
C and the abelian extensionFor almost
all
the group isisomorphic
toZ/2fZ
xJust like in the rational case, we want those
primes
p that havesplitting
field
Bk
inside and for which the order of the Frobenius elementsover p in
Gal(Bk(ql/lk
+ divides the order of the Frobeniuselements over
p in
By
the Chebotarevpartition
argument,
we findagain
that for a’generic’
prime
~,
a fraction£/(£3 -1)
ofthe
primes
p that are inert in O violates(3.4).
Here’generic’
means that tis odd and that has
degree
2t3k(t - 1)
1. UnderGRH,
one canagain
deduce that the inertdensity
6inert (q,
r)
equals
a rational constantc-
times the infinite Eulerproduct occurring
in(3.6)
and(3.8).
In
general,
there are various subtleties that need to be taken care of inthe
analysis
above for £ =2,
when2-power
rootsof q
and r areadjoined
toK or
K(~2n).
We do not go into them in this paper. In theexample
in thenext
section,
we deal with thesecomplications by combining
asimple
adhoc
argument
for a few ’bad’ f with the standard treatment for the’good’
~.4. The
Lagarias example
We now treat the
explicit example
of the modified Lucas sequence which is thesubject
of the theorem stated in the introduction. The roots of the characteristicpolynomial
X2 - X -
1 of the recurrence are 6- =1+2
andits
conjugate
6" =1-2V5.
The initial values xo = 3 and xl = 1yield
aninitial
quotient q
=1-3E
of the sequence. As 7rll = 1 - 3e E 0 =Z ~e~
hasnorm -11,
we find that we have to solve Problem 2 forWe set 1 I as in the
previous
section.
Proof.
As K =Q( J5)
is thequadratic
subfield of(a(~5),
the field hasdegree
2cp(ij)
overQ
if 5 does notdivide ij
anddegree
cp(ij)
if it does.As q
=
is aquotient
of two non-associateprime
elements in 0 and7rl 1
E is a fundamental unit in
C,
thepolynomials
and r areirreducible in
K[X]
for alli, j
EZ>1
by
a standard result as[6,
TheoremVI.9. 1].
Moreover,
the extension K Cgenerated
by
a zero ofXi_q
is
totally
ramified at theprimes
of Klying
over11,
whereas the extensionK C
generated by
a zero ofXV -
r is unramified above 11. Itfollows that K C
K(qlli,
is ofdegree
i2j
for alli, j
EZ>l.
The intersection is contained in the maximal abelian subfield
Ko
ofK (ql/i , rl/ij),
whichequals
One
trivially
computes
Ko
n and the lemma follows. 0 We will need thepreceding
lemmaonly
forsquarefree
j.
In this case, wesimply
have t =#~d
E{5} : dlijl
for odd i.If we substitute the
explicit
degrees
from Lemma 4.1 in(3.7),
we findthat the
split
density
for ourexample equals
where is as in Lemma 4.1. If we set
then the
expression
above may be rewritten asIt is
elementary
to show[9,
Theorem4.2]
thatS’m~n
is the rationalmultiple
- A ~ 0
of the universal constant
arithmetic now
yields
the valuefor the
density
(under GRH)
of theprimes
p - fl mod 5dividing
the78498
primes
below106
aresplit
in K and divide our sequence: a fractionclose to .2601.
For the inert
primes
ofK,
whichsatisfy
p - ±2 mod5,
we have a closerlook at the ’bad’
primes 2,
5 and 11. As for the rationalproblem,
we definethe extensions
... , A--...-,
A_--for
primes
and notethat, by
Lemma4.1,
thefamily
consisting
of the extensionsQ2Q5Qll
and of I~ islinearly independent
over K.Our first observation is that for the inert
primes
p, the order p + 1 of the group Kp in(3.9)
is never divisibleby
5. Condition(3.4)
is therefore automatic for theprime
=5,
and we candisregard
thesplitting
behaviorof p
inSZ5.
We next observe that for inert p, the element r =
-ê2
satisfiesFor
primes
p - 3 mod4,
this shows that(r
modp)2
is the2-Sylow subgroup
of np, so that(3.4)
isagain
automatic for £ = 2. When we nowimpose
thatthe inert
primes
congruent
to 3 mod4,
which form a set ofprimes
ofdensity
1/4,
have the correctsplitting
behavior in the extensionsOf
for £ #
2, 5,
we are
dealing
with alinearly disjoint family
and find(under GRH)
thatthe set of these
primes
hasdensity
We next consider the inert
primes
p - 1 mod 4. For these p, the congruence(4.3)
shows that(r
modp)
has odd order in rp, so(3.4)
is satisfied for £ = 2if and
only
if the orderof q
=-7rfi /1 1
inrp is also odd. As rp is a
cyclic
group of order p + 1 - 2 mod
4,
the orderof q
=(q
modp)
is odd if andonly if q
is a square in ~p. Let x EO/pO
be a square root ofq.
If x is inr,p,
i. e.,
if x has norm 1 inFp,
then its trace x +1/x
is inFp,
and we findthat
is a square modulo p. If x is not in ~p, then x has norm -1 in
Fp
andis a square modulo p. As 5 is not a square modulo our inert
prime
p, wededuce
(4.4) (q
modp)
has odd order in r~P T# - 1 1 is a square modulo p.If p satisfies the
equivalent
conditions of(4.4),
then p is a square modulocase,
(3.4)
is satisfied for £ =2, 5
and 11.Thus,
the set of inertprimes
p - 1 mod 4
satisfying
thequadratic
condition(4.4)
is a set ofprimes
ofdensity
1/8,
and the subset of those p that have the correctsplitting
behavior in the extensions
Qt
for £ #
2, 5,11
has(under GRH)
density
Adding
the fractions obtained for the inertprimes
congruent
to 3 mod 4 and to 1 mod4,
we obtainNumerically,
one finds that 24781primes
out of the 78498primes
below106
are inert in K and divide our sequence: a fraction close to .3157. The sumbplit
+ðinert
is the valuementioned in the theorem in the introduction.
References
[1] C. BALLOT, Density of prime divisors of linear recurrent sequences. Mem. of the AMS 551,
1995.
[2] H. HASSE, Über die Dichte der Primzahlen p, für die eine vorgegebene rationale Zahl a ~ 0
von durch eine vorgegebene Primzahl l ~ 2 teilbarer bzw. unteilbarer Ordnung mod p ist.
Math. Ann. 162 (1965), 74-76.
[3] H. HASSE, Über die Dichte der Primzahlen p, für die eine vorgegebene ganzrationale Zahl
a ~ 0 von gerader bzw. ungerader Ordnung mod p ist. Math. Ann. 166 (1966), 19-23.
[4] C. HOOLEY, On Artin’s conjecture. J. Reine Angew. Math. 225 (1967), 209-220.
[5] J.C. LAGARIAS, The set of primes dividing the Lucas numbers has density 2/3. Pacific J. Math. 118 (1985), 449-461; Errata Ibid. 162 (1994), 393-397.
[6] S. LANG, Algebra, 3rd edition. Addison-Wesley, 1993.
[7] H.W. LENSTRA, JR, On Artin’s conjecture and Euclid’s algorithm in global fields. Inv. Math.
42 (1977), 201-224.
[8] P. MOREE, P. STEVENHAGEN, Prime divisors of Lucas sequences. Acta Arith. 82 (1997),
403-410.
[9] P. MOREE, P. STEVENHAGEN, A two variable Artin conjecture. J. Number Theory 85 (2000),
291-304.
[10] P.J. STEPHENS, Prime divisors of second order linear recurrences. J. Number Theory 8
(1976), 313-332.
[11] P. STEVENHAGEN, Prime densities for second order torsion sequences. preprint (2000).
Pieter MOREE
Korteweg-de Vries Institute for Mathematics Plantage Muidergracht 24 1018 TV Amsterdam The Netherlands moreeovins - uva - nl Peter STEVENHAGEN Mathematisch Instituut
Universiteit Leiden, P.O. Box 9512 2300 RA Leiden
The Netherlands